working paper 153 forecast with judgment and models · forecast with judgment and models by...

Working Paper Research

Forecast with judgment and models

by Francesca Monti

December 2008 No 153

NBB WORKING PAPER No. 153 - DECEMBER 2008

Editorial Director

Jan Smets, Member of the Board of Directors of the National Bank of Belgium

Statement of purpose:

The purpose of these working papers is to promote the circulation of research results (Research Series) and analyticalstudies (Documents Series) made within the National Bank of Belgium or presented by external economists in seminars,conferences and conventions organised by the Bank. The aim is therefore to provide a platform for discussion. The opinionsexpressed are strictly those of the authors and do not necessarily reflect the views of the National Bank of Belgium.

Orders

For orders and information on subscriptions and reductions: National Bank of Belgium,Documentation - Publications service, boulevard de Berlaimont 14, 1000 Brussels

Tel +32 2 221 20 33 - Fax +32 2 21 30 42

The Working Papers are available on the website of the Bank: http://www.nbb.be

© National Bank of Belgium, Brussels

All rights reserved.Reproduction for educational and non-commercial purposes is permitted provided that the source is acknowledged.

ISSN: 1375-680X (print)ISSN: 1784-2476 (online)

http://www.nbb.be/


Abstract

This paper proposes a simple and model-consistent method for combining forecasts generated by

structural micro-founded models and judgmental forecasts. The method also enables the

judgmental forecasts to be interpreted through the lens of the model. We illustrate the proposed

methodology with a real-time forecasting exercise, using a simple neo-Keynesian dynamic

stochastic general equilibrium model and prediction from the Survey of Professional Forecasters.

JEL-code : C32, C53.

Key-words: forecasting, judgment, structural models, Kalman Filter, real time.

Corresponding author:

Francesca Monti, ECARES, Université Libre de Bruxelles (e-mail: [email protected]).

I am greatly indebted to Domenico Giannone, Lucrezia Reichlin and Philippe Weil for invaluableguidance and advice. Finally, I am grateful to Gunter Coenen, David De Antonio Liedo, Marco DelNegro, Andrea Ferrero, Mark Gertler, Paulo Santos Monteiro, Argia Sbordone, Frank Schorfheide,Andrea Tambalotti, Dan Waggoner, Raf Wouters, Tao Zha and all seminar participants at the ECB,University of Pennsylvania, Federal Reserve Bank of NY, Federal Reserve Bank of Atlanta andFederal Reserve Bank of Philadelphia for comments and useful discussions.

The paper was written during my internship in the National Bank of Belgium and I gratefullyacknowledge the financial support of the NBB.

The views expressed in this paper are those of the author and do not necessarily reflect the viewsof the National Bank of Belgium.

mailto:[email protected])

NBB WORKING PAPER - No. 153 - DECEMBER 2008

TABLE OF CONTENTS

1. Introduction..........................................................................................................................1

2. The econometric methodology .............................................................................................4

2.1 The framework ..................................................................................................................4

2.2 Model of the judgmental forecasts......................................................................................6

2.3 Using the model to interpret judgmental forecasts ............................................................13

3. An application ....................................................................................................................13

4. Forecasting and structural analysis ....................................................................................22

5. Conclusions and extensions...............................................................................................32

References ..................................................................................................................................32

National Bank of Belgium - Working papers series .......................................................................35

1 Introduction

Much of the macroeconometric literature of the last decade has focused on

making micro-founded dynamic stochastic general equilibrium (DSGE) mod-

els a viable option for policy analysis and forecasting. Since Smets and

Wouters (2004) have shown that DSGE models estimated with Bayesian

techniques seem to perform quite well in forecasting relative to standard

benchmark models such as VARs, DSGE models have become an increas-

ingly important tool for policy analysis and forecasting at central banks.

Despite their growing use in practice, model-based forecasts still seem

to be outperformed at short horizons -and particularly in the nowcast1- by

forecasts produced by institutional and professional forecasters, such as the

Federal Reserve’s Greenbook (e.g. Sims, 2003) or the Survey of Professional

Forecasters. Where does the advantage of the judgmental forecasters, as I

will define the institutional and professional forecasters from now on, come

from?

Judgmental forecasters monitor and analyze literally hundreds of data

series, using informal methods to distill information from the available data.

Not only they access various data series released by the statistical agencies

(as for example, GDP or industrial production), they also gather other infor-

mation, such as the quantity of goods transported by railway in each month

(Bruno and Lupi, 2004), the electricity consumed each month (Marchetti

and Parigi, 1998) and survey evidence. Moreover, judgmental forecasters are

able to incorporate new data and new information as it becomes available

1Nowcasts are estimates of the current value of variables, such as GDP, that are un-known in the current period due to information lags.

1

throughout the month or the quarter and therefore can take advantage of

the timeliness of this information. Indeed, as Giannone, Reichlin and Small

(2008) point out, timely information seems to play a very important role

in improving the quality of the forecasts, and of the nowcasts in particular.

Finally, in their forecasts, judgmental forecasters account also for all sheerly

judgmental information. A typical example is the adjustments of the fore-

casts made in 1999 in order to account for the fear of the Y2K bug. At the

time this seemed a very important event, but since it had never happened no

model could be expected to encompass it, while the judgmental forecasters

could.

Hence, judgment - i.e. information, knowledge and views outside the

scope of a particular model2 - strongly informs the judgmental forecasts.

The empirical evidence at hand suggests that the ability to account for more,

more timely and ”softer” information is what makes the judgmental forecasts

better at nowcasting and forecasting short horizons.

The introduction of DSGE models in a policy and projection environ-

ment has given rise to a literature on how the model’s outcomes should be

combined with judgmental input and off-model information. The aim of

this paper is to propose a method for combining judgmental forecasts and

model-based forecasts, in order to make predictions that are more accurate

but nevertheless disciplined by rigorous economic theory. In particular, we

propose to interpret the judgmental forecasts as an estimate of the real sig-

nal, which is made with a different, possibly richer, information set and can

be filtered in order to extract the information it contains. Modelling the

2This definition appears in Svensson (2005).

2

judgmental forecasts in the context of the theoretical model also enables us

to interpret them in the light of that same model.

Recently, other authors have addressed the issue of how to use soft data

and judgment in models. Svensson (2005), Svensson and Tetlow (2005) and

Svensson and Williams (2005) develop different frameworks that allow ac-

counting for central bank judgment when constructing optimal policy pro-

jections of the target variables and the instrument rate. They show that such

monetary policy may perform substantially better than monetary policy that

disregards judgment and follows a given instrument rule. Our approach dif-

fers quite substantially from theirs: our goal is solely to produce model-based

forecasts that can account for judgmental and off-model information. Our

approach leaves the structure of the DSGE model unchanged while combining

the model-based forecasts with the judgmental forecasts.

In a Bayesian framework, Robertson, Tallman and Whiteman (2005) sug-

gest a minimum relative entropy procedure for imposing moment restrictions

on simulated forecasts distributions from a variety of models. This technique

involves changing the initial predictive distribution to a new one that sat-

isfies specific moment conditions that come from outside of the models, i.e.

that are judgmental. Therefore, minimum-entropy methods allow adjusting

the full posterior distribution of the DSGE models to match a given experts’

assessment.

Finally, in a joint paper with Giannone and Reichlin (2008), we allow for

the timely information to enter the model directly, not pre-processed by the

judgmental forecasters. In particular, we show how to combine reduced form

estimates of current quarter macroeconomic variables based on a large panel

3

of monthly information with structural micro-founded models which focus on

few key macroeconomic variables (such as GDP, consumption, investment,

inflation). The paper differentiates itself from the emergent literature on

DSGE in data-rich environments (Boivin and Giannoni, 2005), in that it

captures the importance of timely information, allowing for the use of data

with different frequencies and with non-synchronous releases.

This paper is structured as follows. In Section 2 we outline the frame-

work and describe the proposed methodology; we illustrate how to extract

the weights given to the model-based and the judgmental forecast; and de-

scribe how to structuralize the professional forecasts. In Section 3 we apply

the proposed methodology on a prototypical new-Keynesian model using the

Survey of Professional Forecasters’ forecasts to extract judgmental informa-

tion. Section 4 presents the results of the empirical application described

in the previous section. In Section 5 we give some conclusions and outline

future extensions of this paper.

2 The Econometric Methodology

2.1 The Framework

Linear or linearized rational expectations models, solved with methods sug-

gested by Blanchard and Kahn (1980) and Sims (2002) among others, allow

4

a representation for zt in the state space form

st+1 = A(θ)st +B(θ)εt+1 (1)

zt = C(θ)st + νt

where st is an n×1 vector of possibly unobserved state variables, zt is a k×1

vector of variables observed by an econometrician, and εt is an m× 1 vector

of economic shocks impinging on the states, such as shocks to preferences,

technologies, agents’ information sets, and νt is an l×1 vector of measurement

errors (0 6 l 6 k, hence measurement error can be absent or affect some or

all of the variables). A(θ), B(θ) and C(θ) are functions of the underlying

structural parameters of the DSGE model. The εt’s are Gaussian vector

white noise satisfying E(εt) = 0, E(εtε′t) = I, E(εtε

′t+j) = 0, for j > 0.

The νt’s are Gaussian vector white noise satisfying E(νt) = 0, E(νtν′t) = R,

E(νtν′t+j) = 0, for j > 0. The assumption of normality is for convenience

and allows us to associate linear least squares predictions with conditional

expectations. For notational simplicity we will drop the indication that the

matrices A, B, etc. are function of the structural parameters θ.

Associated with the state space representation (1) is the innovations rep-

5

resentation3

st|t = Ast−1|t−1 +Ktut (2)

zt = CAst−1|t−1 + ut

where st|t = E[st|zt, zt−1, ..., z0] is the estimate of the state vector st based on

the observations of zτ up to date t, ut = zt − zt|t−1 = zt − E[zt|zt−1, ..., z0] is

the forecast error made when forecasting zt given the observations of zτ up

to date t-1,

Kt = Ωt|t−1C′(CΩt|t−1C

′ +R)−1,

and Ωt|t−1 = E(st − st|t−1)(st − st|t−1)′ converges to Ω, the unique positive

semidefinite solution that satisfies the algebraic Riccati equation

Ω = BB′ + AΩA′ − AΩC ′(CΩC ′)−1CΩA′. (3)

2.2 Model of the Judgmental Forecasts

The goal of this section is to show how to incorporate judgmental forecasts

into models of the form (1) or, equivalently, (2). In order to do so, we

need to somehow formalize these forecasts. More specifically, we need to

make assumptions on the model and the information set that the judgmental

forecasters use to generate their forecasters.

3The conditions for the existence of this representation are stated carefully, amongothers, in Anderson, Hansen, McGrattan, and Sargent (1996). The conditions are that(A,B,C) be such that iterations on the Riccati equation for Ωt|t−1 = E(st − st|t−1)(st −st|t−1)

′ converge, which makes the associated Kalman gain converge. Sufficient conditionsare that (A, C′) is stabilizable and that (A′, B′) is detectable. See Anderson, Hansen,McGrattan, and Sargent (1996, p. 175) for definitions of stabilizable and detectable.

6

The assumptions about the information available in each period t are

outlined in Table 1. Shocks hit the economy at the beginning of period t.

There are two types of forecasters. The first type generates his forecasts

solely on the basis of the model of the economy (1) and the data released

by the statistical agency. The statistical agency releases data at the end

of each period, so, e.g., in period t there is data available only up to t-1.

Therefore the information set available to the first type of forecaster at time

t comprises exclusively information up to time t-1 : his information set is

It−1 = Spzt−1, zt−2, ...z0, i.e. the space spanned by zt−1, zt−2, ...z0. From

now on I will call the first type of forecaster ’purely’ model-based forecaster.

The second type of forecaster uses a reduced form version of the model

of the economy to make its forecasts - i.e. he or she knows A, B, C, but not

θ - and accesses the information set J , which comprises It−1 but is possibly

more informative. This type represents the judgmental forecasters (JF from

now on). As highlighted in the Introduction, their information set is plausi-

bly richer than It−1: they collect intra-period extra-model information, such

as business surveys, monthly electricity consumption and quantity of goods

transported by railway in each month. This allows them to make a better

estimate of the current value of the variables of interest. Finally, in what

follows, we also assume that J ⊆ It. This means that, once the observable

variables are actually observed, the informational content of the judgmen-

tal forecasts is nihil. Hence, the information the JF can observe does not

improve their estimate of past values of the state variable.

Let us formalize rigorously the judgmental forecasters. At any given time

7

Table 1: Information structure

t t+ 1

shocks hit the economyagents observe them

JF collect informationand release their forecast

stat. agency releasesdata on period t

t their information set J comprises It−1 but is such that, for h > 0

E [utJ′] 6= 0, (4)

ut+h ⊥ J

For τ = 1, 2, ..., t− 1 both purely model-based forecasters and JF are going

to construct the innovations representation (2).4 For τ > t, judgmental

forecasters will report: for h=0,1,...,4

zt+h|J = E[zt+h|J ] + ξht (5)

where E[zt+h|J ] is the least squares forecast made by the JF with their in-

formation set J and ξht is a white noise error that is orthogonal to all the

rest of the information. I.e. we are assuming that the judgmental forecasters

use the reduced form version of model (1) and their richer information set

J to generate the forecasts, but we also allow for the presence of a second

term, which is orthogonal to the rest of the forecasts and can, for example,

be interpreted as a typo made by the forecasters while communicating their

forecasts. ξht , for h=0,1,...,4, has mean zero and variance Σhξ . We allow for

cross-correlation among the elements of ξht (hence Σhξ need not be diagonal),

4This comes from the assumption that J ⊆ It.

8

and covariance matrix

ΣQ ΣS

Σ′S ΣR

,

where

ΣQ = KtEt−1

[utu

′t

]K ′t (7)

ΣS =

[

KtEt−1

[utE [ut|J ]′

]KtEt−1

[utE [ut|J ]′

]K ′tA

′C ′ · · · KtEt−1

[utE [ut|J ]′

]K ′tA

4′C ′

]

ΣR =

V 0U Et−1

[E [ut|J ]E [ut|J ]′

]K ′tA

′C ′ · · · Et−1

[E [ut|J ] E [ut|J ]′

]K ′tA

4′C ′

· V 1U · · · CAKtEt−1

[E [ut|J ] E [ut|J ]′

]K ′tA

4′C ′

.... . .

...

· · · · V 4U

V iU for i=0,1,...,4 is the variance-covariance matrix of the the elements of Ut

associated to zt+i|J .

We obtain E [st|J ] with a 2-step procedure5: first we estimate the matrices

ΣQ, ΣS and ΣR, then we determine E [st|J ] as the expected value of the states

conditional on all past information and today’s judgmental forecasts. In what

follows, we describe the way we obtain estimates for ΣQ, ΣS and ΣR.

First of all, ΣQ can be easily obtained simply using the fact that

E(utu′t) = CΩt|t−1C

′ +R

5Notice that formally we are determining E[st|t|J

]. However, thanks to the law of

iterated expectations, the following holds:

E[st|t|J

]= E [E[s|It]|J ] = E [st|J ] .

10

where Ωt|t−1 = E(st − st|t−1)(st − st|t−1)′. Then notice that

E[

ut(zt|J − zt|t−1

)′]

= E[utE(ut|J)′].

This equality derives from the fact that ξ0t ⊥ ut by assumption. The ut’s and

the zt|t−1’s are readily available from the Kalman filter, so we are able to re-

cover empirically the value of E[ut(zt|J − zt|t−1

)′], and therefore ofE[utE(ut|J)′].

Hence, we can determine the various components of ΣS simply by multiplying

E[ut(zt|J − zt|t−1

)′] by the appropriate matrices, as indicated in (7).

Finally, notice that E(ut|J) is a linear projection of ut on the space

spanned by J , i.e.

ut = E(ut|J) + µt

where µt is orthogonal to the space spanned by J . Therefore,

E[utE(ut|J)′] = E[E(ut|J)E(ut|J)′], (8)

i.e. the variance of the expected value of the current period innovation

given the information set J , E(ut|J) is equal to the covariance among the

innovation and its expected value. Hence we are able to pin down the value

of all the off-diagonal components of ΣR. The diagonal blocks of ΣR are

the covariance matrices of the single elements of Ut in equation (6) and can

obtained empirically simply as the covariance matrices of(zt+h|J − zt+h|t−1

),

for h = 0, 1, ..., 4.

The matrices ΣQ, ΣS and ΣR are estimated using the information that is

available at each point in time, i.e. up to t-1. The knowledge of the matrices

11

ΣS and ΣR enables us to determine E [st|J ] simply as:

E [st|J ] = Ast−1|t−1 + ΣSΣ−1

R

(zt|J − zt|t−1

)

...(zt+4|J − zt+4|t−1

)

︸︷︷︸

Ut

. (9)

The augmented forecasts are then straightforward to derive:

z+

t+h|J = CAhE [st|J ] (10)

for h = 0, 1, ..., 4.

The forecasts in (10) combine the judgemental forecasts with the model-

based forecasts in a non trivial manner, because they use the forecasts of

all horizons provided by the SPF, i.e. up to one year ahead. The weights

assigned to the judgemental forecasts in generating E [st|J ] are determined

by the matrices ΣS and ΣR. ΣS, the covariance of the innovations and the

difference between the judgmental forecasts and the model, is a measure of

how much information the judgmental forecasters have on the the current

state of the economy. ΣR is a measure of the volatility of their forecasts and

depends also on the variance Σhξ . The bigger is ΣS , the more weight is given

to the judgmental forecasts in determining the new estimate of the state;

a greater ΣR, instead, causes the judgmental forecasts to be down-weighed.

These matrices are re-estimated (recursively or with a rolling window) every

period using information up to t-1 and are, hence, a measure of the JF past

performance.

12

2.3 Using the model to interpret judgemental forecasts

Another interesting aspect of this procedure is that it also allows interpret-

ing the judgmental forecasts through the lens of the model. Storytelling

is difficult when it comes to judgmental forecasts; in our set-up we will be

able to interpret the forecasts in light of the model and therefore somehow

structuralize the forecasts.

We would like to recover the estimates of the structural shocks made by

the judgmental forecasters. Notice that if we take the expectation of Bεt (in

the state equation of (1)) given the information set J, we obtain

E [Bεt|J ] = E [st|J ] −E [Ast−1|J ]

= E [st|J ] −Ast−1|t−1

= ΣSΣ−1

R Ut.

The equalities above hold because of the law of iterated expectations and

because of our assumption that the information the JF can observe does not

improve their estimate of past values of the state variable, i.e. It−1 ⊆ J ⊆ It.

Then, if B has the appropriate dimensions, we can extract the JFs’ estimate

of the structural shock by simply multiplying ΣSΣ−1

R Ut by (B′B)−1B′.

3 An application

We use a simple new-keynesian dynamic stochastic general equilibrium model,

as the one used in Del Negro and Schorfheide (2004). The model consists of a

representative household, a continuum of monopolistically competitive firms

13

and monetary policy authority that sets the nominal interest rate in response

to deviations of inflation and output from their targets. The representative

household derives disutility from hours worked and utility from consumption

C relative to a habit stock and real money balances MP

. We assume that

the habit stock is given by the level of technology A.6 The representative

household maximizes expected utility

Et

[∞∑

s=t

βs−t(

(Cs/As)1−τ − 1

1 − τ+ χ log

Ms

Ps− hs

)]

(11)

where β is the discount factor, τ the risk aversion parameter and χ is a scale

factor. P is the economy-wide nominal price level that the household takes

as given. The (gross) inflation rate is defined as πt = Pt

Pt−1.

The household supplies perfectly elastic labor supply services to the firm

period by period and receives in return real wage W. It also has access to a

domestic capital market on which they can trade nominal government bonds

B that pay gross interest rate R. Moreover, the household receives aggre-

gate residual profits D and has to pay lump-sum taxes T. Hence, its budget

constraint is:

Ct +Bt

Pt+Mt

Pt+TtPt

= Wtht +Mt−1

Pt+Rt−1

Bt−1

Pt+Dt (12)

The transversality condition on asset accumulation rules out Ponzi schemes.

On the production side, there is a continuum of monopolistically com-

petitive firms, each facing a downward-sloping demand curve, derived in the

6This assumption ensures that the economy evolves along a balanced growth path.

14

usual way from Dixit-Stiglitz type of preferences, for its differentiated prod-

uct

Pt(j) =

(Xt(j)

Xt

)−1/ν

Pt, (13)

where Pt(j) is the profit-maximizing price that is consistent with production

level Xt(j), while Pt is the aggregate price level and Xt is aggregate demand

(both beyond the control of the individual firm). The parameter ν is the

elasticity of substitution between two differentiated goods. We assume that

the firms face quadratic adjustment costs: that is, when a firm wants to

change its price beyond the economy-wide inflation rate π∗, it incurs menu

costs in terms of lost output:

ACt(j) =φ

2

(Pt(j)

Pt−1(j)− π∗

)2

Xt(j). (14)

The presence of these adjustment costs determines the presence of nominal

rigidities, and the parameter φ > 0 determines the degree of stickiness within

the economy.

The production function is linear in labor, which is hired from the house-

hold:

Xt(j) = Atht(j). (15)

Total factor productivity At follows a unit root process of the form:

lnAt = ln γ + lnAt−1 + zt, (16)

15

where

zt = ρz zt−1 + εz,t. (17)

Hence, there will be a stochastic trend in the model. εz,t can be broadly

interpreted as a technology shock that affects all firms in the same way.

The maximization problem faced by the firm is the following:

maxEt

[∞∑

s=t

QsDs(j)

]

(18)

subject to (15) and (16), and where the j-th firm’s profit Ds(j) is

Ds(j) =

(

Ps(j)

PsXs(j) −Wshs(j) −

φ

2

(Ps(j)

Ps−1(j)− π∗

)2

Xs(j)

)

. (19)

Qs is the time-dependent discount factor that firms use to evaluate future

profit streams. Although firms are heterogeneous ex-ante, we only consider

the symmetric equilibrium in which all firms behave identically and can be

aggregated into a single representative monopolistically competitive firm.

Since the household is the recipient of the firms’ residual payments, it di-

rects firms to make decisions based on the household’s intertemporal rate of

substitution. Hence Qt+1/Qt = β(Ct/Ct+1)τ .

The monetary policy authority follows an interest rate rule, such that it

adjusts its instruments in response to deviations of inflation and output from

their respective targets:

Rt

R∗=

(Rt−1

R∗

)ρR

[(πtπ∗

)ψ1

(Xt

X∗t

)ψ2

]1−ρR

eεR,t (20)

16

where R∗ is the steady-state nominal interest rate andX∗t is potential output,

which we defined as X∗t = At after normalizing hours worked to one. The

central bank supplies the money demanded by the households to support the

desired nominal interest rate. The parameter 0 6 ρR < 1 governs the degree

of interest rate smoothing, while εR,t can be interpreted as an unanticipated

deviation from the policy rule.

The government consumes a fraction ζt of each individual good and levies

a lump-sum tax (or subsidy) Tt/Pt to finance any shortfall in government

revenues (or to rebate any surplus), so its budget constraint is:

ζtXt +Rt−1

Bt−1

Pt+Mt−1

Pt=Bt

Pt+Bt

Pt+Mt

Pt. (21)

The fiscal authority accomodates the monetary policy of the central bank

and endogenously adjusts the primary surplus to changes in the government’s

outstanding liabilities. Finally, we define gt = 1/(1 − ζt) and assume that

gt = ln(gt/g∗) follows a stationary AR(1) process

gt = ρg gt−1 + εg,t (22)

where εg,t can be broadly interpreted as a government spending shock.

To solve the model, we derive the optimality conditions from the maxi-

mization problem. Consumption, output, wages and the marginal utility of

consumption are detrended by the total factor productivity At, in order to

obtain a model that has a deterministic steady-state in terms of the detrended

17

variables. The loglinearized system can be reduced to:

yt = Etyt+1 +1

τEtπt+1 −

1

τrt + Etgt+1 +

ρzτzt

πt = βEtπt+1 + κ(yt − gt)

rt = ψ1(1 − ρr)πt + ψ2(1 − ρr)yt + ρr rt−1 + 0.25εr,t (23)

gt = ρggt−1 + εg,t

zt = ρz zt−1 + εz,t,

where β = eγ−r∗/4

100 . The relation between logdeviations from steady state and

observable output growth, GDP deflator inflation and the annual nominal

interest rate is given by the following measurement equation:

∆ lnYt = ln γ + yt − yt−1 + zt

INFLt = π∗ + 4πt (24)

RAt = π∗ + r∗ + 4rt

The model given by equations (23) and (24) can then solved with standard

techniques, such as those proposed by Blanchard and Kahn (1980) and Sims

(2002), among others, and hence cast in a standard state-space model (1).

In what follows, we will perform an out-of-sample real-time forecasting

exercise, using as evaluation sample the period 1992-2006. We use real-time

quarterly data for real GDP, GDP deflator inflation and the Fed Funds rate

for the US, which is available from the Philadelphia Fed’s website, and choose

as a starting point for our data the first quarter of 1982.7 The Bayesian es-

7Due to the unavailability of real-time data on population, we have made the some-

18

Prior Distribution Posterior DistributionDistribution mean st.dev mode mean st.dev

γ Normal 0.5 0.5 0.6815 0.6924 0.1251π∗ Gamma 5 2 3.8862 4.3852 1.3813r∗ Gamma 2 1 3.0139 3.0022 0.5148τ Gamma 2 0.5 2.8826 2.9628 0.5015κ Gamma 0.3 0.1 0.1188 0.1528 0.0500ψ1 Gamma 1.5 0.5 1.0369 1.5420 0.3651ψ2 Gamma 0.125 0.1 0.0951 0.2719 0.0320ρg Beta 0.9 0.05 0.9656 0.9648 0.0200ρz Beta 0.2 0.1 0.3244 0.3265 0.1107ρr Beta 0.7 0.15 0.8048 0.8279 0.0359σg InvGamma 1.25 0.65 0.4924 0.5066 0.0706σz InvGamma 1.25 0.65 0.5927 0.6404 0.0754σr InvGamma 0.63 0.33 0.7155 0.7310 0.0794

Table 2: Prior and posterior distribution of the parameters of the modelestimated over the period 1982Q1 to 1995Q4.

timation of the model’s parameters is performed recursively every two years.

The forecasting exercise is totally in real time: hence, when forecasting (0

to 4 steps ahead), e.g., in 1996Q1, we will use only data available at that

vintage. As an example, Table 2 reports the estimate for the model’s pa-

rameters made in 1996Q1. We use the Survey of Professional Forecasters

(SPF) as example of judgmental forecast. The Survey of Professional Fore-

casters, conducted by the Federal Reserve Bank of Philadelphia, is based on

many individual commercial and academic forecasts, which are then grouped

in mean or median forecasts. The Survey is conducted near the end of the

second month of each quarter and publishes forecasts for the current quarter

and the next 4 quarters in the future. The forecasts for real GDP are avail-

what heroic assumption that the population has been constant throughout the periodconsidered.

19

able only from 1981Q3 on. Notice that the SPF does not provide forecasts

for the Fed Funds rate.

An important data-related issue regards the appropriate ”actual” series

to use when comparing the various forecasts. Because macroeconomic data is

continuously revised, we need to make a choice about which revision to use.

Following Romer and Romer (2000), we choose to use the second revision,

i.e. the one done at the end of the subsequent quarter. The second revision

seems to be the appropriate series to use because it is based on relatively

complete data, but it is still roughly contemporaneous with the forecasts

we are analyzing. This series does not include rebenchmarking and changes

in the definition of the economic concepts that occur in the annual and

quinquennial revisions and should, therefore, be conceptually similar to the

series being forecast.

Let us now present some forecasting results that will highlight the mo-

tivation of this paper. Throughout the paper we will compare the forecasts

we produced with the forecasts of naive benchmark models. The benchmark

model for GDP in levels is a random walk with drift (hence GDP growth

in this model is simply constant). We estimate this constant as the mean

growth of GDP in the 10 years previous to the date at which we perform the

evaluation. The benchmark models for GDP deflator inflation and the Fed

Funds rate are random walks, i.e. the forecast of the value of inflation and

the interest in the next period is today’s value.

The first two columns of table 3 report the out-of-sample performance of

the forecasts generated with the simple new-keynesian model and of the SPF

relative to the naive benchmark, for GDP growth, GDP deflator inflation

20

and the Fed Funds rate respectively. In the first column of each table is

reported, for each variable of interest, the ratio of the mean square error of the

purely model-based forecast (NK) against the mean square error of the naive

benchmark, while the second column reports the ratio of the mean square

error of the SPF against the mean square error of the naive benchmark.

Asterisks indicate a rejection of the test of equal predictive accuracy between

each forecast and the naive benchmark.8

This simple NK model does not have much forecasting power, but it has

the advantage of enabling us to tell a consistent economic story about the

forecasts. The professional forecasters fare better than the model, in that

their forecasts, at least at 0 and 1 step ahead, are better than the naive

benchmark in a statistical sense. And this holds both for GDP growth and

for GDP deflator growth. Over time, model and judgement seem to contain

information that is useful in different points in time (see Figure 2): in few

periods the model does even better than the SPF. Many methods, such as

optimal-weight combination forecasts (e.g., Stock and Watson, 2004), have

been developed to take advantage of the difference in performance through

time. The available methods however are statistical, reduced-form methods

and therefore do not allow telling compelling stories about the forecasts. Our

methodology instead delivers both features.

In the following section we present the results obtained when applying the

8Following Romer and Romer (2000), our inference is based on the regression: (zht −zm

ht)2 − (zht − znaive

ht)2 = c + uht where z is the variable to be forecasted at horizon h

using model -m . The estimate of c is simply the difference between forecast-m and aNaive model MSFEs, and the standard error is corrected for heteroskedasticity and serialcorrelation over h-1 months. This testing procedure falls in the Diebold-Mariano-Westframework.

21

methodology proposed in Section 2 to the model presented above and using

the SPF forecasts. We will show that the methodology we propose allows

for generating new augmented model-based forecasts that are more accurate

than the original ’purely’ model-based ones, but that can still reflect the

economic stories that arise from the model. We will compare the our forecasts

not only against forecasts produced by a naive benchmark, but also against

another forecast obtained by combining model and SPF in a more standard

way. In particular, we will use an equal-weights combination of the ’purely’

model-based forecasts and the SPF, which, as many authors (e.g. Stock and

Watson, 2004) point out, generally outperforms not only the single forecasts

but also more sophisticated and time-varying optimal combinations.

4 Forecasting and Structural Analysis

In this section we present the results obtained applying the proposed

methodology to the framework described in the previous section. First, we

present model-based forecasts for real GDP, the GDP deflator and the Fed

Funds rate that can account for the judgmental information contained in

the SPF forecasts. We will compare their performance on the basis of their

mean square forecast error, i.e. deeming better a forecast with a smaller

MSE9. Second, we will discuss the way the SPF and the ’purely’ model-

based forecasts are combined, showing the weights associated to the SPF in

generating estimates of the underlying states and how they change in time.

9The results hold if performance is measured differently, e.g. with Mean AbsoluteErrors.

22

GDP growth forecasts relative to constant growthEVALUATION SAMPLE: 1992:2 - 2006:4

Horizon NK SPF COMB AUGMQ0 1.2320 0.8199 ∗ 0.9275 0.7579 ∗Q1 1.1680 0.9349 0.9930 1.0608Q2 1.1022 1.0002 1.0160 1.0739Q3 1.0547 0.9642 0.9819 1.0529Q4 1.0389 1.0006 0.9972 1.0400

GDP deflator inflation relative to a random walkEVALUATION SAMPLE: 1996:2 - 2006:4

Horizon NK SPF COMB AUGMQ0 0.9383 0.6365 ∗∗ 0.6495 ∗∗ 0.5599 ∗∗Q1 0.9275 0.6577 ∗∗ 0.6743∗∗ 0.6433 ∗∗Q2 1.0204 0.8286 0.7887 0.9319Q3 1.1158 0.9862 0.8920 1.0963Q4 1.0992 0.9055 0.8781 1.0147

Fed Funds rate relative to a random walkEVALUATION SAMPLE: 1996:2 - 2006:4

Horizon NK SPF COMB AUGMQ0 2.4154 NaN NaN 2.4166Q1 2.0412 NaN NaN 2.0991Q2 1.7235 NaN NaN 1.7854Q3 1.4957 NaN NaN 1.5583Q4 1.3336 NaN NaN 1.3958

Table 3: Relative MSFE of forecasts of GDP growth, GDP deflator inflationand the fed funds rate with respect to their naive benchmarks. Asterisks de-note forecasts that are statistically more accurate than the naive benchmarkat 1% (∗ ∗ ∗), 5%(∗∗) and 10%(∗)

23

Finally, we will show how the method we propose also allows to read the

judgmental forecasts through the lens of the model, and hence enables us to

interpret these forecasts.

We construct the augmented forecasts (AUGM) as described in the previ-

ous section. The evaluation sample goes from the first quarter of 1992 to the

fourth quarter of 2006. The matrices ΣS and ΣR are estimated recursively

every quarter, i.e. at every new release of the SPF. As noted before, the SPF

does not provide forecasts of the Fed Funds rate, so we treat this missing ob-

servation problem as suggested in Giannone, Reichlin and Small (2008)10.

We also compare the augmented forecasts with forecasts produced combin-

ing the ’purely’ model-based and the SPF forecasts with equal weights. The

latter are identified in the tables and graphs as COMB.

Table 3 reports the mean square forecast error (MSE) of the purely model-

based forecasts (NK), the SPF, the combination forecasts and the augmented

forecasts relative to the naive model11, when forecasting GDP growth. The

augmented forecasts outperform all the other nowcasts, including the com-

bination nowcast. At higher horizons, instead, the augmented forecasts con-

verge to the model-based forecasts and hence perform less well (though still

better than the model). The reason is that we have assumed that the infor-

mation set of the judgmental forecasters contains information on the only on

the current, and not on the future, innovations. Therefore, the augmented

forecasts assign virtually no weight to the SPF for higher horizon forecasts12.

10Assigning infinite variance to the noise term when observations are missing, one im-plicitely sets to zero the weight that is assigned that variable in the filtering problem.

11The benchmark model for GDP in levels is a random walk with drift, while benchmarkmodels for inflation and the Fed Funds rate are random walk.

12If we made the assumption that the SPF might have information also on future shocks,

24

Q3−92 Q1−95 Q3−97 Q1−00 Q3−02 Q1−05

0

0.5

1

1.5

GDP growth

GDPNKSPFAUGMCOMB

Q3−92 Q1−95 Q3−97 Q1−00 Q3−02 Q1−05

0

0.5

1

1.5

2

2.5

3

GDP deflator Inflation

DGDPNKSPFAUGMCOMB

Figure 1: Nowcast for GDP growth (top panel) and GDP deflatorinflation (bottom panel) - EVALUATION SAMPLE 1992:1-2006:4.The grey thick solid line represents the actual data, the dotted line is the ’purely’model-based nowcast, the line with the x-marker is the SPF, the dashed line is thecombination forecast and the thick line with the circle marker is the augmentedforecast.

25

For this reason we will mainly focus our analysis on the nowcast.

Figure 1 reports the nowcasts of the model (NK), of the SPF, of the com-

bination model (COMB) and of the augmented model (AUGM) compared

with actual data for GDP growth and annualized quarter-on-quarter-inflation

over the full evaluation sample. The augmented forecasts of both variables

track more closely the model-based forecasts in the periods in which the

model performs better than the SPF, such as in the second half of the 90’s,

while it mimics the SPF when they fare better than the model.

Figure 2 reports the smoothed forecast errors for the nowcast of GDP and

GDP deflator inflation (centered moving average 4 quarters on each side) over

the full sample period. The dotted line is the ’purely’ model-based nowcast,

the line with the x-marker is the SPF, the dashed line is the combination

forecast and the thick line with the circle marker is the augmented forecast.

These figures confirm that the augmented nowcasts are consistently more

accurate than the naive benchmark and that the augmented are able track

more closely the ’purely’ model-based forecast or the SPF, depending on

their respective past performance.

The way these two forecasts are combined is not trivial. The SPF forecasts

then the augmented forecasts would be incorporated also at higher horizons. We do notwant to go there, however, because it would put at risk the theoretical consistency of theforecasts. Indeed, the fact that the SPF have more information than the agents, but thelatter do not account for that is at odds with the assumption of the rationality of theagents.

26

Q1−94 Q1−96 Q1−98 Q1−00 Q1−02 Q1−04

0.5

1

1.5

2

2.5

GDP growth

SPFNKmodelCOMBAUG

Q1−94 Q1−96 Q1−98 Q1−00 Q1−02 Q1−04

0.5

1

1.5

2

GDP deflator Inflation

SPFNKCOMBAUG

Figure 2: NOWCASTS: Smoothed Square forecast errors for GDPgrowth and GDP deflator inflation. This figure reports the smoothed(centered moving average 4 quarters on each side) forecast errors for thenowcast of GDP growth (top panel) and of GDP deflator inflation (bottompanel) over the full sample period. The dotted line is the ’purely’ model-based nowcast, the line with the x-marker is the SPF, the dashed line isthe combination forecast and the thick line with the circle marker is theaugmented forecast.

27

Q1−92 Q1−97 Q1−02 Q4−06−2

−1

0

1

2

GDPgrowtht

r t

Q1−92 Q1−97 Q1−02 Q4−06−2

−1

0

1

2

r t

INFLt

Q1−92 Q1−97 Q1−02 Q4−06−2

−1

0

1

2

r t

FFRt

Q1−92 Q1−97 Q1−02 Q4−06−2

−1

0

1

2

gt

GDPgrowtht

Q1−92 Q1−97 Q1−02 Q4−06−2

−1

0

1

2

gt

INFLt

Q1−92 Q1−97 Q1−02 Q4−06−2

−1

0

1

2

gt

FFRt

Q1−92 Q1−97 Q1−02 Q4−06−2

−1

0

1

2

zt

GDPgrowtht

Q1−92 Q1−97 Q1−02 Q4−06−2

−1

0

1

2

zt

INFLt

Q1−92 Q1−97 Q1−02 Q4−06−2

−1

0

1

2

zt

FFRt

Q0 Q1 Q2 Q3 Q4

Figure 3: Weights assigned to the SPF forecasts when estimatingthe states of the model - from 1992:Q1 to 2006Q4. Each panel ofthis figure shows the weight assigned, throughout the evaluation sample, tothe SPF forecasts at different horizons. The columns report the role playedby the forecasts of a specific variable (e.g. GDP growth in the first column)in determining the estimate of the different states. The rows indicated thecontribution of the SPF forecast of each variable for one specific state (e.g.rt in the first row).

28

are used to generate new estimates of the state variables as follows:

z+

t|J = C

Ast−1|t−1 + ΣSΣ−1

R

(zt|J − zt|t−1

)

...(zt+4|J − zt+4|t−1

)

If the SPF forecasts carry a lot of information about a state, they will have a

big weight in the determination of E[st|J ]. Figure (3) reports these weights:

each plot of the figure shows the weight assigned, throughout the evaluation

sample, to the SPF forecasts at different horizons (i.e. plot contains 5 lines,

one for each forecasting horizon). Each of the three rows of plots shows

the contribution of the SPF forecasts of GDP growth, inflation and the Fed

Funds rate in the estimation of a specific state (rt, gt and zt respectively). The

columns show the weights assigned to the SPF forecasts of a specific variable

in determining the states. The last column, for example, is full of zeros,

because the SPF do not produce a forecast of the Fed Funds rate and hence

it does not contribute to the estimation of the states. Let us now analyze

the figure along the rows. The first row of graphs shows the contribution

of the SPF forecasts of GDP growth, inflation and the Fed Funds rate in

determining the augmented estimate of the state rt, while the second and

third rows of graphs show the contribution of the SPF forecasts of GDP

growth, inflation and the Fed Funds rate in determining the new estimate of

the states gt and zt. The figure suggests that the SPF forecasts carry very

little information on rt, but are instead very useful for gt and zt. Obviously,

given our assumptions forecasts with shorter horizons are attributed more

29

weight; moreover these weights are higher in the first part of the evaluation

sample, because back then the performance of the SPF had been consistently

better than the model. Finally, we can infer that, in the estimation of zt (gt),

most of, but not all, the weight is given to the SPF forecasts of GDP growth

(Inflation).

Finally, we present an example of ”structural” analysis of the SPF fore-

casts. Figure 4 reports the ex-post estimates of the technology shock and

the government spending shock (thick grey line) and the structural shocks

as perceived by the SPF (black with the x-shaped markers). The latter are

constructed as described in section 2, while the former are simply obtained

premutliplying the innovations deriving from the Kalman smoother on the

dataset 1982Q1-2006Q4 by (B′B)−1B′. Of course, these are not necessarily

the real structural shocks, just the ex-post estimates, hence the best esti-

mates available.

To see how this analysis can be useful, think, for example, of the widely

discussed question: why did the most professional forecasters miss out on the

real activity boom of the 90’s? Figure 4 provides a possible answer: indeed,

it seem that they underestimated the technology shocks that were happen-

ing in that period. Another episode that is often discussed is the delay with

which the professional and institutional forecasters called the 2001 recession.

Our decomposition provides a story for this: apparently the forecasters un-

derestimated the negative technology shocks of the first quarters of 2001,

while overestimating the negative demand shock in the wake of 9/11.

30

Q3−92 Q1−95 Q3−97 Q1−00 Q3−02 Q1−05

−1

−0.5

0

0.5

Technology shock

Q3−92 Q1−95 Q3−97 Q1−00 Q3−02 Q1−05

−0.5

0

0.5

Government spending shock

ex−post estimates SPF estimates

Figure 4: Ex-post estimates of the technology shock (top panel) andgovernment spending shock (bottom panel) and these shocks asperceived by the SPF when nowcasting 1992Q1-2006Q4. The greythick line represents the ex-post estimates of shock, while the dotted line with thetriangle-shaped markers is the shock as perceived by the SPF.

31

5 Conclusions and Extensions

In this paper I propose a model-consistent and parsimonious method of com-

bining judgmental and model-based forecasts. I suggest modeling the judg-

mental forecasts as optimal estimates of the variables of interest, made with

a different, possibly more informative, information set. I then show how

they can be accounted for in the framework of a linearized and solved DSGE

model. The methodology I propose allows generating forecasts that are more

accurate than the purely model-based ones, but that are still disciplined by

the economic rigor of the model.

I have also highlighted how to infer the information content of the judg-

mental forecasts from the weights that the augmented forecasts assign to

them. More precisely, the more the professional forecasters are able to gather

information on the shocks, the more the augmented forecast will use the pro-

fessional forecasts when combining them with the predictions from the model,

but it will down-weigh them if the variance of their forecast errors is too large.

Finally I have described how to interpret the forecasts through the lens of

the model, by extracting the structural shocks as they are perceived by the

professional forecasters. This can give interesting answers to widely debated

questions, as shown in the example.

References

[1] Anderson, E., L. P. Hansen, E. R. McGrattan and T. J. Sargent (1996):“Mechanics of Forming and Estimating Dynamic Linear Economies,” in

32

Handbook of Computational Economics, Volume 1, ed. by D. A. K. HansM. Amman, and J. Rust, pp. 171–252. North-Holland

[2] Blanchard, O.J. and C.M. Kahn (1980), ”The Solution of Linear Differ-ence Models under Rational Expectations,”, Econometrica 48(5), 1305-1311

[3] Boivin, J. and M. Giannoni (2005), ”DGSE in a Data-Rich Environ-ment”

[4] Bruno, G. and C. Lupi (2004), ”Forecasting industrial production andthe early detection of turning points”, Empirical Economics, 29, 647-671

[5] Del Negro, M. and F. Schorfheide (2004). ”Priors from General Equilib-rium Models for VARS,” International Economic Review, Departmentof Economics, University of Pennsylvania and Osaka University Instituteof Social and Economic Research Association, vol. 45(2), 643-673

[6] Diebold, F.X. and R.S. Mariano (1995),”Comparing Predictive Accu-racy,” Journal of Business Economics and Statistics, 13, 253-265

[7] Giannone, D., F. Monti and L. Reichlin (2008), ”Incorporating Con-junctural Analysis in Structural Models,” ECARES mimeo

[8] Giannone, D., Reichlin, L. and Small, D., (2008), ”Nowcasting: Thereal-time informational content of macroeconomic data,” Journal ofMonetary Economics 55(4), 665-676

[9] Marchetti, D.J. and G. Parigi (1998), ”Energy Consumption, SurveyData and the Prediction of Industrial Production in Italy”, Temi diDiscussione del Servizio Studi della Banca d’ Italia, No 342

[10] , Robertson, J., E. Tallman and C. Whiteman (2005), ”Forecasting usingRelative Entropy,” Journal of Money Credit and Banking 37, 383-402.

[11] Romer, C. and D. Romer (2000), ”Federal Reserve Information and theBehavior of Interest Rates”, The American Economic Review, 90, 3,429-457

[12] Smets, F. and R. Wouters (2004), ”Forecasting with a Bayesian DSGEmodel - an application to the euro area,” Working Paper Series 389,European Central Bank

[13] Sims, C.A (2002), ”Solving Linear Rational Expectations Models,”Computational Economics, Springer, 20(1-2), 1-20.

33

[14] Sims, C.A (2003), ”The role of models and probabilities in the monetarypolicy process,” Brooking Papers on Economic Activity, 2002:2, 1-63

[15] Svensson, L.E.O (2005), ”Monetary Policy with Judgement: ForecastTargeting”, NBER Working Paper 11167

[16] Stock, J. and M. Watson (2004), ”Combination Forecasts of OutputGrowth in a Seven-Country Data Set”, Journal of Forecasting 23, 405-430

[17] Svensson, L.E.O and R.J. Tetlow (2005), ”Optimal Policy Projections”,NBER Working Paper 11392

[18] Svensson, L.E.O. and N. Williams (2005), ”Monetary Policy with ModelUncertainty: Distribution Forecast Targeting”, NBER Working Paper11733

34

NBB WORKING PAPER No. 153 - DECEMBER 2008 35

NATIONAL BANK OF BELGIUM - WORKING PAPERS SERIES

1. "Model-based inflation forecasts and monetary policy rules" by M. Dombrecht and R. Wouters, ResearchSeries, February 2000.

2. "The use of robust estimators as measures of core inflation" by L. Aucremanne, Research Series,February 2000.

3. "Performances économiques des Etats-Unis dans les années nonante" by A. Nyssens, P. Butzen,P. Bisciari, Document Series, March 2000.

4. "A model with explicit expectations for Belgium" by P. Jeanfils, Research Series, March 2000.5. "Growth in an open economy: some recent developments" by S. Turnovsky, Research Series, May 2000.6. "Knowledge, technology and economic growth: an OECD perspective" by I. Visco, A. Bassanini,

S. Scarpetta, Research Series, May 2000.7. "Fiscal policy and growth in the context of European integration" by P. Masson, Research Series, May

2000.8. "Economic growth and the labour market: Europe's challenge" by C. Wyplosz, Research Series, May

2000.9. "The role of the exchange rate in economic growth: a euro-zone perspective" by R. MacDonald,

Research Series, May 2000.10. "Monetary union and economic growth" by J. Vickers, Research Series, May 2000.11. "Politique monétaire et prix des actifs: le cas des Etats-Unis" by Q. Wibaut, Document Series, August

2000.12. "The Belgian industrial confidence indicator: leading indicator of economic activity in the euro area?" by

J.-J. Vanhaelen, L. Dresse, J. De Mulder, Document Series, November 2000.13. "Le financement des entreprises par capital-risque" by C. Rigo, Document Series, February 2001.14. "La nouvelle économie" by P. Bisciari, Document Series, March 2001.15. "De kostprijs van bankkredieten" by A. Bruggeman and R. Wouters, Document Series, April 2001.16. "A guided tour of the world of rational expectations models and optimal policies" by Ph. Jeanfils,

Research Series, May 2001.17. "Attractive Prices and Euro - Rounding effects on inflation" by L. Aucremanne and D. Cornille,

Documents Series, November 2001.18. "The interest rate and credit channels in Belgium: an investigation with micro-level firm data" by

P. Butzen, C. Fuss and Ph. Vermeulen, Research series, December 2001.19. "Openness, imperfect exchange rate pass-through and monetary policy" by F. Smets and R. Wouters,

Research series, March 2002.20. "Inflation, relative prices and nominal rigidities" by L. Aucremanne, G. Brys, M. Hubert, P. J. Rousseeuw

and A. Struyf, Research series, April 2002.21. "Lifting the burden: fundamental tax reform and economic growth" by D. Jorgenson, Research series,

May 2002.22. "What do we know about investment under uncertainty?" by L. Trigeorgis, Research series, May 2002.23. "Investment, uncertainty and irreversibility: evidence from Belgian accounting data" by D. Cassimon,

P.-J. Engelen, H. Meersman, M. Van Wouwe, Research series, May 2002.24. "The impact of uncertainty on investment plans" by P. Butzen, C. Fuss, Ph. Vermeulen, Research series,

May 2002.25. "Investment, protection, ownership, and the cost of capital" by Ch. P. Himmelberg, R. G. Hubbard,

I. Love, Research series, May 2002.26. "Finance, uncertainty and investment: assessing the gains and losses of a generalised non-linear

structural approach using Belgian panel data", by M. Gérard, F. Verschueren, Research series,May 2002.

27. "Capital structure, firm liquidity and growth" by R. Anderson, Research series, May 2002.28. "Structural modelling of investment and financial constraints: where do we stand?" by J.- B. Chatelain,

Research series, May 2002.29. "Financing and investment interdependencies in unquoted Belgian companies: the role of venture

capital" by S. Manigart, K. Baeyens, I. Verschueren, Research series, May 2002.30. "Development path and capital structure of Belgian biotechnology firms" by V. Bastin, A. Corhay,

G. Hübner, P.-A. Michel, Research series, May 2002.31. "Governance as a source of managerial discipline" by J. Franks, Research series, May 2002.


32. "Financing constraints, fixed capital and R&D investment decisions of Belgian firms" by M. Cincera,Research series, May 2002.

33. "Investment, R&D and liquidity constraints: a corporate governance approach to the Belgian evidence"by P. Van Cayseele, Research series, May 2002.

34. "On the Origins of the Franco-German EMU Controversies" by I. Maes, Research series, July 2002.35. "An estimated dynamic stochastic general equilibrium model of the Euro Area", by F. Smets and

R. Wouters, Research series, October 2002.36. "The labour market and fiscal impact of labour tax reductions: The case of reduction of employers' social

security contributions under a wage norm regime with automatic price indexing of wages", byK. Burggraeve and Ph. Du Caju, Research series, March 2003.

37. "Scope of asymmetries in the Euro Area", by S. Ide and Ph. Moës, Document series, March 2003.38. "De autonijverheid in België: Het belang van het toeleveringsnetwerk rond de assemblage van

personenauto's", by F. Coppens and G. van Gastel, Document series, June 2003.39. "La consommation privée en Belgique", by B. Eugène, Ph. Jeanfils and B. Robert, Document series,

June 2003.40. "The process of European monetary integration: a comparison of the Belgian and Italian approaches", by

I. Maes and L. Quaglia, Research series, August 2003.41. "Stock market valuation in the United States", by P. Bisciari, A. Durré and A. Nyssens, Document series,

November 2003.42. "Modeling the Term Structure of Interest Rates: Where Do We Stand?, by K. Maes, Research series,

February 2004.43. "Interbank Exposures: An Empirical Examination of System Risk in the Belgian Banking System", by

H. Degryse and G. Nguyen, Research series, March 2004.44. "How Frequently do Prices change? Evidence Based on the Micro Data Underlying the Belgian CPI", by

L. Aucremanne and E. Dhyne, Research series, April 2004.45. "Firms' investment decisions in response to demand and price uncertainty", by C. Fuss and

Ph. Vermeulen, Research series, April 2004.46. "SMEs and Bank Lending Relationships: the Impact of Mergers", by H. Degryse, N. Masschelein and

J. Mitchell, Research series, May 2004.47. "The Determinants of Pass-Through of Market Conditions to Bank Retail Interest Rates in Belgium", by

F. De Graeve, O. De Jonghe and R. Vander Vennet, Research series, May 2004.48. "Sectoral vs. country diversification benefits and downside risk", by M. Emiris, Research series,

May 2004.49. "How does liquidity react to stress periods in a limit order market?", by H. Beltran, A. Durré and P. Giot,

Research series, May 2004.50. "Financial consolidation and liquidity: prudential regulation and/or competition policy?", by

P. Van Cayseele, Research series, May 2004.51. "Basel II and Operational Risk: Implications for risk measurement and management in the financial

sector", by A. Chapelle, Y. Crama, G. Hübner and J.-P. Peters, Research series, May 2004.52. "The Efficiency and Stability of Banks and Markets", by F. Allen, Research series, May 2004.53. "Does Financial Liberalization Spur Growth?" by G. Bekaert, C.R. Harvey and C. Lundblad, Research

series, May 2004.54. "Regulating Financial Conglomerates", by X. Freixas, G. Lóránth, A.D. Morrison and H.S. Shin, Research

series, May 2004.55. "Liquidity and Financial Market Stability", by M. O'Hara, Research series, May 2004.56. "Economisch belang van de Vlaamse zeehavens: verslag 2002", by F. Lagneaux, Document series,

June 2004.57. "Determinants of Euro Term Structure of Credit Spreads", by A. Van Landschoot, Research series,

July 2004.58. "Macroeconomic and Monetary Policy-Making at the European Commission, from the Rome Treaties to

the Hague Summit", by I. Maes, Research series, July 2004.59. "Liberalisation of Network Industries: Is Electricity an Exception to the Rule?", by F. Coppens and

D. Vivet, Document series, September 2004.60. "Forecasting with a Bayesian DSGE model: an application to the euro area", by F. Smets and

R. Wouters, Research series, September 2004.61. "Comparing shocks and frictions in US and Euro Area Business Cycle: a Bayesian DSGE approach", by

F. Smets and R. Wouters, Research series, October 2004.


62. "Voting on Pensions: A Survey", by G. de Walque, Research series, October 2004.63. "Asymmetric Growth and Inflation Developments in the Acceding Countries: A New Assessment", by

S. Ide and P. Moës, Research series, October 2004.64. "Importance économique du Port Autonome de Liège: rapport 2002", by F. Lagneaux, Document series,

November 2004.65. "Price-setting behaviour in Belgium: what can be learned from an ad hoc survey", by L. Aucremanne and

M. Druant, Research series, March 2005.66. "Time-dependent versus State-dependent Pricing: A Panel Data Approach to the Determinants of

Belgian Consumer Price Changes", by L. Aucremanne and E. Dhyne, Research series, April 2005.67. "Indirect effects – A formal definition and degrees of dependency as an alternative to technical

coefficients", by F. Coppens, Research series, May 2005.68. "Noname – A new quarterly model for Belgium", by Ph. Jeanfils and K. Burggraeve, Research series,

May 2005.69. "Economic importance of the Flemish maritime ports: report 2003", F. Lagneaux, Document series, May

2005.70. "Measuring inflation persistence: a structural time series approach", M. Dossche and G. Everaert,

Research series, June 2005.71. "Financial intermediation theory and implications for the sources of value in structured finance markets",

J. Mitchell, Document series, July 2005.72. "Liquidity risk in securities settlement", J. Devriese and J. Mitchell, Research series, July 2005.73. "An international analysis of earnings, stock prices and bond yields", A. Durré and P. Giot, Research

series, September 2005.74. "Price setting in the euro area: Some stylized facts from Individual Consumer Price Data", E. Dhyne,

L. J. Álvarez, H. Le Bihan, G. Veronese, D. Dias, J. Hoffmann, N. Jonker, P. Lünnemann, F. Rumler andJ. Vilmunen, Research series, September 2005.

75. "Importance économique du Port Autonome de Liège: rapport 2003", by F. Lagneaux, Document series,October 2005.

76. "The pricing behaviour of firms in the euro area: new survey evidence, by S. Fabiani, M. Druant,I. Hernando, C. Kwapil, B. Landau, C. Loupias, F. Martins, T. Mathä, R. Sabbatini, H. Stahl andA. Stokman, Research series, November 2005.

77. "Income uncertainty and aggregate consumption, by L. Pozzi, Research series, November 2005.78. "Crédits aux particuliers - Analyse des données de la Centrale des Crédits aux Particuliers", by

H. De Doncker, Document series, January 2006.79. "Is there a difference between solicited and unsolicited bank ratings and, if so, why?" by P. Van Roy,

Research series, February 2006.80. "A generalised dynamic factor model for the Belgian economy - Useful business cycle indicators and

GDP growth forecasts", by Ch. Van Nieuwenhuyze, Research series, February 2006.81. "Réduction linéaire de cotisations patronales à la sécurité sociale et financement alternatif" by

Ph. Jeanfils, L. Van Meensel, Ph. Du Caju, Y. Saks, K. Buysse and K. Van Cauter, Document series,March 2006.

82. "The patterns and determinants of price setting in the Belgian industry" by D. Cornille and M. Dossche,Research series, May 2006.

83. "A multi-factor model for the valuation and risk management of demand deposits" by H. Dewachter,M. Lyrio and K. Maes, Research series, May 2006.

84. "The single European electricity market: A long road to convergence", by F. Coppens and D. Vivet,Document series, May 2006.

85. "Firm-specific production factors in a DSGE model with Taylor price setting", by G. de Walque, F. Smetsand R. Wouters, Research series, June 2006.

86. "Economic importance of the Belgian ports: Flemish maritime ports and Liège port complex - report2004", by F. Lagneaux, Document series, June 2006.

87. "The response of firms' investment and financing to adverse cash flow shocks: the role of bankrelationships", by C. Fuss and Ph. Vermeulen, Research series, July 2006.

88. "The term structure of interest rates in a DSGE model", by M. Emiris, Research series, July 2006.89. "The production function approach to the Belgian output gap, Estimation of a Multivariate Structural Time

Series Model", by Ph. Moës, Research series, September 2006.90. "Industry Wage Differentials, Unobserved Ability, and Rent-Sharing: Evidence from Matched Worker-

Firm Data, 1995-2002", by R. Plasman, F. Rycx and I. Tojerow, Research series, October 2006.


91. "The dynamics of trade and competition", by N. Chen, J. Imbs and A. Scott, Research series, October2006.

92. "A New Keynesian Model with Unemployment", by O. Blanchard and J. Gali, Research series, October2006.

93. "Price and Wage Setting in an Integrating Europe: Firm Level Evidence", by F. Abraham, J. Konings andS. Vanormelingen, Research series, October 2006.

94. "Simulation, estimation and welfare implications of monetary policies in a 3-country NOEM model", byJ. Plasmans, T. Michalak and J. Fornero, Research series, October 2006.

95. "Inflation persistence and price-setting behaviour in the euro area: a summary of the Inflation PersistenceNetwork evidence ", by F. Altissimo, M. Ehrmann and F. Smets, Research series, October 2006.

96. "How Wages Change: Micro Evidence from the International Wage Flexibility Project", by W.T. Dickens,L. Goette, E.L. Groshen, S. Holden, J. Messina, M.E. Schweitzer, J. Turunen and M. Ward, Researchseries, October 2006.

97. "Nominal wage rigidities in a new Keynesian model with frictional unemployment", by V. Bodart,G. de Walque, O. Pierrard, H.R. Sneessens and R. Wouters, Research series, October 2006.

98. "Dynamics on monetary policy in a fair wage model of the business cycle", by D. De la Croix,G. de Walque and R. Wouters, Research series, October 2006.

99. "The kinked demand curve and price rigidity: evidence from scanner data", by M. Dossche, F. Heylenand D. Van den Poel, Research series, October 2006.

100. "Lumpy price adjustments: a microeconometric analysis", by E. Dhyne, C. Fuss, H. Peseran andP. Sevestre, Research series, October 2006.

101. "Reasons for wage rigidity in Germany", by W. Franz and F. Pfeiffer, Research series, October 2006.102. "Fiscal sustainability indicators and policy design in the face of ageing", by G. Langenus, Research

series, October 2006.103. "Macroeconomic fluctuations and firm entry: theory and evidence", by V. Lewis, Research series,

October 2006.104. "Exploring the CDS-Bond Basis" by J. De Wit, Research series, November 2006.105. "Sector Concentration in Loan Portfolios and Economic Capital", by K. Düllmann and N. Masschelein,

Research series, November 2006.106. "R&D in the Belgian Pharmaceutical Sector", by H. De Doncker, Document series, December 2006.107. "Importance et évolution des investissements directs en Belgique", by Ch. Piette, Document series,

January 2007.108. "Investment-Specific Technology Shocks and Labor Market Frictions", by R. De Bock, Research series,

February 2007.109. "Shocks and frictions in US Business cycles: a Bayesian DSGE Approach", by F. Smets and R. Wouters,

Research series, February 2007.110. "Economic impact of port activity: a disaggregate analysis. The case of Antwerp", by F. Coppens,

F. Lagneaux, H. Meersman, N. Sellekaerts, E. Van de Voorde, G. van Gastel, Th. Vanelslander,A. Verhetsel, Document series, February 2007.

111. "Price setting in the euro area: some stylised facts from individual producer price data", byPh. Vermeulen, D. Dias, M. Dossche, E. Gautier, I. Hernando, R. Sabbatini, H. Stahl, Research series,March 2007.

112. "Assessing the Gap between Observed and Perceived Inflation in the Euro Area: Is the Credibility of theHICP at Stake?", by L. Aucremanne, M. Collin, Th. Stragier, Research series, April 2007.

113. "The spread of Keynesian economics: a comparison of the Belgian and Italian experiences", by I. Maes,Research series, April 2007.

114. "Imports and Exports at the Level of the Firm: Evidence from Belgium", by M. Muûls and M. Pisu,Research series, May 2007.

115. "Economic importance of the Belgian ports: Flemish maritime ports and Liège port complex - report2005", by F. Lagneaux, Document series, May 2007.

116. "Temporal Distribution of Price Changes: Staggering in the Large and Synchronization in the Small", byE. Dhyne and J. Konieczny, Research series, June 2007.

117. "Can excess liquidity signal an asset price boom?", by A. Bruggeman, Research series, August 2007.118. "The performance of credit rating systems in the assessment of collateral used in Eurosystem monetary

policy operations", by F. Coppens, F. González and G. Winkler, Research series, September 2007.119. "The determinants of stock and bond return comovements", by L. Baele, G. Bekaert and K. Inghelbrecht,

Research series, October 2007.


120. "Monitoring pro-cyclicality under the capital requirements directive: preliminary concepts for developing aframework", by N. Masschelein, Document series, October 2007.

121. "Dynamic order submission strategies with competition between a dealer market and a crossingnetwork", by H. Degryse, M. Van Achter and G. Wuyts, Research series, November 2007.

122. "The gas chain: influence of its specificities on the liberalisation process", by C. Swartenbroekx,Document series, November 2007.

123. "Failure prediction models: performance, disagreements, and internal rating systems", by J. Mitchell andP. Van Roy, Research series, December 2007.

124. "Downward wage rigidity for different workers and firms: an evaluation for Belgium using the IWFPprocedure", by Ph. Du Caju, C. Fuss and L. Wintr, Research series, December 2007.

125. "Economic importance of Belgian transport logistics", by F. Lagneaux, Document series, January 2008.126. "Some evidence on late bidding in eBay auctions", by L. Wintr, Research series, January 2008.127. "How do firms adjust their wage bill in Belgium? A decomposition along the intensive and extensive

margins", by C. Fuss, Research series, January 2008.128. "Exports and productivity – comparable evidence for 14 countries", by The International Study Group on

Exports and Productivity, Research series, February 2008.129. "Estimation of monetary policy preferences in a forward-looking model: a Bayesian approach", by

P. Ilbas, Research series, March 2008.130. "Job creation, job destruction and firms' international trade involvement", by M. Pisu, Research series,

March 2008.131. "Do survey indicators let us see the business cycle? A frequency decomposition", by L. Dresse and

Ch. Van Nieuwenhuyze, Research series, March 2008.132. "Searching for additional sources of inflation persistence: the micro-price panel data approach", by

R. Raciborski, Research series, April 2008.133. "Short-term forecasting of GDP using large monthly datasets - A pseudo real-time forecast evaluation

exercise", by K. Barhoumi, S. Benk, R. Cristadoro, A. Den Reijer, A. Jakaitiene, P. Jelonek, A. Rua,G. Rünstler, K. Ruth and Ch. Van Nieuwenhuyze, Research series, June 2008.

134. "Economic importance of the Belgian ports: Flemish maritime ports, Liège port complex and the port ofBrussels - report 2006" by S. Vennix, Document series, June 2008.

135. "Imperfect exchange rate pass-through: the role of distribution services and variable demand elasticity",by Ph. Jeanfils, Research series, August 2008.

136. "Multivariate structural time series models with dual cycles: Implications for measurement of output gapand potential growth", by Ph. Moës, Research series, August 2008.

137. "Agency problems in structured finance - a case study of European CLOs", by J. Keller, Documentseries, August 2008.

138. "The efficiency frontier as a method for gauging the performance of public expenditure: a Belgian casestudy", by B. Eugène, Research series, September 2008.

139. "Exporters and credit constraints. A firm-level approach", by M. Muûls, Research series, September2008.

140. "Export destinations and learning-by-exporting: Evidence from Belgium", by M. Pisu, Research series,September 2008.

141. "Monetary aggregates and liquidity in a neo-Wicksellian framework", by M. Canzoneri, R. Cumby,B. Diba and D. López-Salido, Research series, October 2008.

142 "Liquidity, inflation and asset prices in a time-varying framework for the euro area, by Ch. Baumeister,E. Durinck and G. Peersman, Research series, October 2008.

143. "The bond premium in a DSGE model with long-run real and nominal risks", by Glenn D. Rudebusch andEric T. Swanson, Research series, October 2008.

144. "Imperfect information, macroeconomic dynamics and the yield curve: an encompassing macro-financemodel", by H. Dewachter, Research series, October 2008.

145. "Housing market spillovers: evidence from an estimated DSGE model", by M. Iacoviello and S. Neri,Research series, October 2008.

146. "Credit frictions and optimal monetary policy", by V. Cúrdia and M. Woodford, Research series,October 2008.

147. "Central Bank misperceptions and the role of money in interest rate rules", by G. Beck and V. Wieland,Research series, October 2008.

148. "Financial (in)stability, supervision and liquidity injections: a dynamic general equilibrium approach", byG. de Walque, O. Pierrard and A. Rouabah, Research series, October 2008.


149. "Monetary policy, asset prices and macroeconomic conditions: a panel-VAR study", byK. Assenmacher-Wesche and S. Gerlach, Research series, October 2008.

150. "Risk premiums and macroeconomic dynamics in a heterogeneous agent model", by F. De Graeve,M. Dossche, M. Emiris, H. Sneessens and R. Wouters, Research series, October 2008.

151. "Financial factors in economic fluctuations", by L. J. Christiano, R. Motto and M. Rotagno, Researchseries, to be published.

152. "Rent-sharing under different bargaining regimes: Evidence from linked employer-employee data" byM. Rusinek and F. Rycx, Research series, December 2008.

153. "Forecast with judgment and models" by F. Monti, Research series, December 2008.